The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/audit_scope.py script executes local git commands to calculate diffs, list commits, and establish the scope of the work being audited.
Evidence: The script calls subprocess.run using a list of arguments (non-shell) within the git(), detect_base(), and main() functions.
Context: These operations are necessary for the skill's primary function and follow security best practices by avoiding shell interpretation.
[PROMPT_INJECTION]: As a code and documentation auditing tool, the skill is designed to ingest and process external content such as git diffs, commit messages, and external coverage reports, which creates a surface for indirect prompt injection where malicious instructions could be embedded in the audited data.
Ingestion points: scripts/audit_scope.py (via git diff, git log, and reading coverage report files).
Boundary markers: Absent. The tool output is processed by the agent without specific delimiters to isolate potentially untrusted content from the agent's control logic.
Capability inventory: scripts/audit_scope.py (local git command execution and file reading).
Sanitization: Present. The script includes a robust reject_entity_declarations function that uses the expat parser to detect and block XML External Entity (XXE) attacks in coverage reports before they are processed by the primary XML library.

self-audit